We once spent the better part of a quarter automating an onboarding process, and we did an excellent job of it. What had taken a person two days of clicking through five systems now took eleven minutes and required no human at all. It was the cleanest piece of automation the team had built that year, and within a month it had created more mess than the manual process ever had, because the process itself was wrong and we had simply removed every last opportunity for a human to notice.
The manual version had been slow, which was its hidden feature. Somewhere in those two days, a person would look at a request and think, that department does not usually need this level of access, let me check. That pause was not in any documentation and it appeared on no process map, but it was doing real quality control. When we automated the steps, we automated the steps we could see, and quietly deleted the judgement that had been holding the whole thing together in the gaps between them.
This is the failure mode I now watch for whenever someone proposes automating something painful. Automation is an amplifier, not a corrective. It takes whatever the process currently produces and produces much more of it, much faster, with far greater consistency. If the process is sound, that is transformative. If the process is subtly broken, you have industrialised the defect and removed the friction that was limiting the blast radius, and the errors now arrive at machine speed with a reassuring green tick beside them.
So the first question is no longer how do we automate this. It is what does this process actually do, including the parts nobody wrote down, and is it correct? Automation makes an excellent second step and a terrible first one. Fix the process while a human is still in the loop to see the results, and only then take the human out, because after that the feedback goes quiet.
Automating a good process is leverage. Automating a bad one is just committing to the same mistake several thousand times a day without having to think about it.
– Serguey Shinder
Top comments (0)